Reinforcement Learning-Enhanced Dynamic Multi-Objective Evolutionary Algorithm for Cooperative Shovel Scheduling in Open-Pit Mining

Aiming at the ore blending problem in the open-pit mining scenario, the optimization objective is to minimize the grade variance of the sliding window of the ore flow, and a cooperative shovel scheduling model for real-time ore blending is established. A reinforcement learning-enhanced dynamic multi- objective evolutionary algorithm is proposed, where a Q-learning operator selection mechanism is introduced to reuse the information from the previous environment for tracking the optimal solution in the dynamic environment. Experimental results show that the proposed method can effectively control the grade fluctuation of the ore flow and dynamically respond to external events, and can find a better solution in real time compared with the traditional operator selection algorithm.